prepagg.c File Reference

#include "postgres.h"
#include "access/htup_details.h"
#include "catalog/pg_aggregate.h"
#include "catalog/pg_type.h"
#include "nodes/nodeFuncs.h"
#include "nodes/pathnodes.h"
#include "optimizer/clauses.h"
#include "optimizer/cost.h"
#include "optimizer/optimizer.h"
#include "optimizer/plancat.h"
#include "optimizer/prep.h"
#include "parser/parse_agg.h"
#include "utils/builtins.h"
#include "utils/datum.h"
#include "utils/fmgroids.h"
#include "utils/lsyscache.h"
#include "utils/memutils.h"
#include "utils/syscache.h"

Include dependency graph for prepagg.c:

Go to the source code of this file.

Functions
static bool	preprocess_aggrefs_walker (Node *node, PlannerInfo *root)
static int	find_compatible_agg (PlannerInfo *root, Aggref *newagg, List **same_input_transnos)
static int	find_compatible_trans (PlannerInfo *root, Aggref *newagg, bool shareable, Oid aggtransfn, Oid aggtranstype, int transtypeLen, bool transtypeByVal, Oid aggcombinefn, Oid aggserialfn, Oid aggdeserialfn, Datum initValue, bool initValueIsNull, List *transnos)
static Datum	GetAggInitVal (Datum textInitVal, Oid transtype)
void	preprocess_aggrefs (PlannerInfo *root, Node *clause)
static void	preprocess_aggref (Aggref *aggref, PlannerInfo *root)
void	get_agg_clause_costs (PlannerInfo *root, AggSplit aggsplit, AggClauseCosts *costs)

Function Documentation

find_compatible_agg()

static int find_compatible_agg ( PlannerInfo *  root, Aggref *  newagg, List **  same_input_transnos  )

static

Definition at line 361 of file prepagg.c.

{
    ListCell   *lc;
    int         aggno;
 
    *same_input_transnos = NIL;
 
    /* we mustn't reuse the aggref if it contains volatile function calls */
    if (contain_volatile_functions((Node *) newagg))
        return -1;
 
    /*
     * Search through the list of already seen aggregates.  If we find an
     * existing identical aggregate call, then we can re-use that one.  While
     * searching, we'll also collect a list of Aggrefs with the same input
     * parameters.  If no matching Aggref is found, the caller can potentially
     * still re-use the transition state of one of them.  (At this stage we
     * just compare the parsetrees; whether different aggregates share the
     * same transition function will be checked later.)
     */
    aggno = -1;
    foreach(lc, root->agginfos)
    {
        AggInfo    *agginfo = lfirst_node(AggInfo, lc);
        Aggref     *existingRef;
 
        aggno++;
 
        existingRef = linitial_node(Aggref, agginfo->aggrefs);
 
        /* all of the following must be the same or it's no match */
        if (newagg->inputcollid != existingRef->inputcollid ||
            newagg->aggtranstype != existingRef->aggtranstype ||
            newagg->aggstar != existingRef->aggstar ||
            newagg->aggvariadic != existingRef->aggvariadic ||
            newagg->aggkind != existingRef->aggkind ||
            !equal(newagg->args, existingRef->args) ||
            !equal(newagg->aggorder, existingRef->aggorder) ||
            !equal(newagg->aggdistinct, existingRef->aggdistinct) ||
            !equal(newagg->aggfilter, existingRef->aggfilter))
            continue;
 
        /* if it's the same aggregate function then report exact match */
        if (newagg->aggfnoid == existingRef->aggfnoid &&
            newagg->aggtype == existingRef->aggtype &&
            newagg->aggcollid == existingRef->aggcollid &&
            equal(newagg->aggdirectargs, existingRef->aggdirectargs))
        {
            list_free(*same_input_transnos);
            *same_input_transnos = NIL;
            return aggno;
        }
 
        /*
         * Not identical, but it had the same inputs.  If the final function
         * permits sharing, return its transno to the caller, in case we can
         * re-use its per-trans state.  (If there's already sharing going on,
         * we might report a transno more than once.  find_compatible_trans is
         * cheap enough that it's not worth spending cycles to avoid that.)
         */
        if (agginfo->shareable)
            *same_input_transnos = lappend_int(*same_input_transnos,
                                               agginfo->transno);
    }
 
    return -1;
}

References Aggref::aggcollid, Aggref::aggdirectargs, Aggref::aggdistinct, Aggref::aggfilter, Aggref::aggfnoid, PlannerInfo::agginfos, Aggref::aggkind, Aggref::aggorder, AggInfo::aggrefs, Aggref::aggstar, Aggref::aggtype, Aggref::aggvariadic, Aggref::args, contain_volatile_functions(), equal(), Aggref::inputcollid, lappend_int(), lfirst_node, linitial_node, list_free(), NIL, AggInfo::shareable, and AggInfo::transno.

Referenced by preprocess_aggref().

find_compatible_trans()

static int find_compatible_trans ( PlannerInfo *  root, Aggref *  newagg, bool  shareable, Oid  aggtransfn, Oid  aggtranstype, int  transtypeLen, bool  transtypeByVal, Oid  aggcombinefn, Oid  aggserialfn, Oid  aggdeserialfn, Datum  initValue, bool  initValueIsNull, List *  transnos  )

static

Definition at line 439 of file prepagg.c.

{
    ListCell   *lc;
 
    /* If this aggregate can't share transition states, give up */
    if (!shareable)
        return -1;
 
    foreach(lc, transnos)
    {
        int         transno = lfirst_int(lc);
        AggTransInfo *pertrans = list_nth_node(AggTransInfo,
                                               root->aggtransinfos,
                                               transno);
 
        /*
         * if the transfns or transition state types are not the same then the
         * state can't be shared.
         */
        if (aggtransfn != pertrans->transfn_oid ||
            aggtranstype != pertrans->aggtranstype)
            continue;
 
        /*
         * The serialization and deserialization functions must match, if
         * present, as we're unable to share the trans state for aggregates
         * which will serialize or deserialize into different formats.
         * Remember that these will be InvalidOid if they're not required for
         * this agg node.
         */
        if (aggserialfn != pertrans->serialfn_oid ||
            aggdeserialfn != pertrans->deserialfn_oid)
            continue;
 
        /*
         * Combine function must also match.  We only care about the combine
         * function with partial aggregates, but it's too early in the
         * planning to know if we will do partial aggregation, so be
         * conservative.
         */
        if (aggcombinefn != pertrans->combinefn_oid)
            continue;
 
        /*
         * Check that the initial condition matches, too.
         */
        if (initValueIsNull && pertrans->initValueIsNull)
            return transno;
 
        if (!initValueIsNull && !pertrans->initValueIsNull &&
            datumIsEqual(initValue, pertrans->initValue,
                         transtypeByVal, transtypeLen))
            return transno;
    }
    return -1;
}

References PlannerInfo::aggtransinfos, AggTransInfo::aggtranstype, AggTransInfo::combinefn_oid, datumIsEqual(), AggTransInfo::deserialfn_oid, initValue(), AggTransInfo::initValueIsNull, lfirst_int, list_nth_node, AggTransInfo::serialfn_oid, and AggTransInfo::transfn_oid.

Referenced by preprocess_aggref().

get_agg_clause_costs()

void get_agg_clause_costs	(	PlannerInfo *	root,
		AggSplit	aggsplit,
		AggClauseCosts *	costs
	)

Definition at line 541 of file prepagg.c.

{
    ListCell   *lc;
 
    foreach(lc, root->aggtransinfos)
    {
        AggTransInfo *transinfo = lfirst_node(AggTransInfo, lc);
 
        /*
         * Add the appropriate component function execution costs to
         * appropriate totals.
         */
        if (DO_AGGSPLIT_COMBINE(aggsplit))
        {
            /* charge for combining previously aggregated states */
            add_function_cost(root, transinfo->combinefn_oid, NULL,
                              &costs->transCost);
        }
        else
            add_function_cost(root, transinfo->transfn_oid, NULL,
                              &costs->transCost);
        if (DO_AGGSPLIT_DESERIALIZE(aggsplit) &&
            OidIsValid(transinfo->deserialfn_oid))
            add_function_cost(root, transinfo->deserialfn_oid, NULL,
                              &costs->transCost);
        if (DO_AGGSPLIT_SERIALIZE(aggsplit) &&
            OidIsValid(transinfo->serialfn_oid))
            add_function_cost(root, transinfo->serialfn_oid, NULL,
                              &costs->finalCost);
 
        /*
         * These costs are incurred only by the initial aggregate node, so we
         * mustn't include them again at upper levels.
         */
        if (!DO_AGGSPLIT_COMBINE(aggsplit))
        {
            /* add the input expressions' cost to per-input-row costs */
            QualCost    argcosts;
 
            cost_qual_eval_node(&argcosts, (Node *) transinfo->args, root);
            costs->transCost.startup += argcosts.startup;
            costs->transCost.per_tuple += argcosts.per_tuple;
 
            /*
             * Add any filter's cost to per-input-row costs.
             *
             * XXX Ideally we should reduce input expression costs according
             * to filter selectivity, but it's not clear it's worth the
             * trouble.
             */
            if (transinfo->aggfilter)
            {
                cost_qual_eval_node(&argcosts, (Node *) transinfo->aggfilter,
                                    root);
                costs->transCost.startup += argcosts.startup;
                costs->transCost.per_tuple += argcosts.per_tuple;
            }
        }
 
        /*
         * If the transition type is pass-by-value then it doesn't add
         * anything to the required size of the hashtable.  If it is
         * pass-by-reference then we have to add the estimated size of the
         * value itself, plus palloc overhead.
         */
        if (!transinfo->transtypeByVal)
        {
            int32       avgwidth;
 
            /* Use average width if aggregate definition gave one */
            if (transinfo->aggtransspace > 0)
                avgwidth = transinfo->aggtransspace;
            else if (transinfo->transfn_oid == F_ARRAY_APPEND)
            {
                /*
                 * If the transition function is array_append(), it'll use an
                 * expanded array as transvalue, which will occupy at least
                 * ALLOCSET_SMALL_INITSIZE and possibly more.  Use that as the
                 * estimate for lack of a better idea.
                 */
                avgwidth = ALLOCSET_SMALL_INITSIZE;
            }
            else
            {
                avgwidth = get_typavgwidth(transinfo->aggtranstype, transinfo->aggtranstypmod);
            }
 
            avgwidth = MAXALIGN(avgwidth);
            costs->transitionSpace += avgwidth + 2 * sizeof(void *);
        }
        else if (transinfo->aggtranstype == INTERNALOID)
        {
            /*
             * INTERNAL transition type is a special case: although INTERNAL
             * is pass-by-value, it's almost certainly being used as a pointer
             * to some large data structure.  The aggregate definition can
             * provide an estimate of the size.  If it doesn't, then we assume
             * ALLOCSET_DEFAULT_INITSIZE, which is a good guess if the data is
             * being kept in a private memory context, as is done by
             * array_agg() for instance.
             */
            if (transinfo->aggtransspace > 0)
                costs->transitionSpace += transinfo->aggtransspace;
            else
                costs->transitionSpace += ALLOCSET_DEFAULT_INITSIZE;
        }
    }
 
    foreach(lc, root->agginfos)
    {
        AggInfo    *agginfo = lfirst_node(AggInfo, lc);
        Aggref     *aggref = linitial_node(Aggref, agginfo->aggrefs);
 
        /*
         * Add the appropriate component function execution costs to
         * appropriate totals.
         */
        if (!DO_AGGSPLIT_SKIPFINAL(aggsplit) &&
            OidIsValid(agginfo->finalfn_oid))
            add_function_cost(root, agginfo->finalfn_oid, NULL,
                              &costs->finalCost);
 
        /*
         * If there are direct arguments, treat their evaluation cost like the
         * cost of the finalfn.
         */
        if (aggref->aggdirectargs)
        {
            QualCost    argcosts;
 
            cost_qual_eval_node(&argcosts, (Node *) aggref->aggdirectargs,
                                root);
            costs->finalCost.startup += argcosts.startup;
            costs->finalCost.per_tuple += argcosts.per_tuple;
        }
    }
}

References add_function_cost(), Aggref::aggdirectargs, AggTransInfo::aggfilter, PlannerInfo::agginfos, AggInfo::aggrefs, PlannerInfo::aggtransinfos, AggTransInfo::aggtransspace, AggTransInfo::aggtranstype, AggTransInfo::aggtranstypmod, ALLOCSET_DEFAULT_INITSIZE, ALLOCSET_SMALL_INITSIZE, AggTransInfo::args, AggTransInfo::combinefn_oid, cost_qual_eval_node(), AggTransInfo::deserialfn_oid, DO_AGGSPLIT_COMBINE, DO_AGGSPLIT_DESERIALIZE, DO_AGGSPLIT_SERIALIZE, DO_AGGSPLIT_SKIPFINAL, AggClauseCosts::finalCost, AggInfo::finalfn_oid, get_typavgwidth(), lfirst_node, linitial_node, MAXALIGN, OidIsValid, QualCost::per_tuple, AggTransInfo::serialfn_oid, QualCost::startup, AggClauseCosts::transCost, AggTransInfo::transfn_oid, AggClauseCosts::transitionSpace, and AggTransInfo::transtypeByVal.

Referenced by create_grouping_paths(), create_partial_grouping_paths(), and estimate_path_cost_size().

GetAggInitVal()

static Datum GetAggInitVal ( Datum  textInitVal, Oid  transtype  )

static

Definition at line 503 of file prepagg.c.

{
    Oid         typinput,
                typioparam;
    char       *strInitVal;
    Datum       initVal;
 
    getTypeInputInfo(transtype, &typinput, &typioparam);
    strInitVal = TextDatumGetCString(textInitVal);
    initVal = OidInputFunctionCall(typinput, strInitVal,
                                   typioparam, -1);
    pfree(strInitVal);
    return initVal;
}

References getTypeInputInfo(), OidInputFunctionCall(), pfree(), and TextDatumGetCString.

Referenced by preprocess_aggref().

preprocess_aggref()

static void preprocess_aggref ( Aggref *  aggref, PlannerInfo *  root  )

static

Definition at line 117 of file prepagg.c.

{
    HeapTuple   aggTuple;
    Form_pg_aggregate aggform;
    Oid         aggtransfn;
    Oid         aggfinalfn;
    Oid         aggcombinefn;
    Oid         aggserialfn;
    Oid         aggdeserialfn;
    Oid         aggtranstype;
    int32       aggtranstypmod;
    int32       aggtransspace;
    bool        shareable;
    int         aggno;
    int         transno;
    List       *same_input_transnos;
    int16       resulttypeLen;
    bool        resulttypeByVal;
    Datum       textInitVal;
    Datum       initValue;
    bool        initValueIsNull;
    bool        transtypeByVal;
    int16       transtypeLen;
    Oid         inputTypes[FUNC_MAX_ARGS];
    int         numArguments;
 
    Assert(aggref->agglevelsup == 0);
 
    /*
     * Fetch info about the aggregate from pg_aggregate.  Note it's correct to
     * ignore the moving-aggregate variant, since what we're concerned with
     * here is aggregates not window functions.
     */
    aggTuple = SearchSysCache1(AGGFNOID,
                               ObjectIdGetDatum(aggref->aggfnoid));
    if (!HeapTupleIsValid(aggTuple))
        elog(ERROR, "cache lookup failed for aggregate %u",
             aggref->aggfnoid);
    aggform = (Form_pg_aggregate) GETSTRUCT(aggTuple);
    aggtransfn = aggform->aggtransfn;
    aggfinalfn = aggform->aggfinalfn;
    aggcombinefn = aggform->aggcombinefn;
    aggserialfn = aggform->aggserialfn;
    aggdeserialfn = aggform->aggdeserialfn;
    aggtranstype = aggform->aggtranstype;
    aggtransspace = aggform->aggtransspace;
 
    /*
     * Resolve the possibly-polymorphic aggregate transition type.
     */
 
    /* extract argument types (ignoring any ORDER BY expressions) */
    numArguments = get_aggregate_argtypes(aggref, inputTypes);
 
    /* resolve actual type of transition state, if polymorphic */
    aggtranstype = resolve_aggregate_transtype(aggref->aggfnoid,
                                               aggtranstype,
                                               inputTypes,
                                               numArguments);
    aggref->aggtranstype = aggtranstype;
 
    /*
     * If transition state is of same type as first aggregated input, assume
     * it's the same typmod (same width) as well.  This works for cases like
     * MAX/MIN and is probably somewhat reasonable otherwise.
     */
    aggtranstypmod = -1;
    if (aggref->args)
    {
        TargetEntry *tle = (TargetEntry *) linitial(aggref->args);
 
        if (aggtranstype == exprType((Node *) tle->expr))
            aggtranstypmod = exprTypmod((Node *) tle->expr);
    }
 
    /*
     * If finalfn is marked read-write, we can't share transition states; but
     * it is okay to share states for AGGMODIFY_SHAREABLE aggs.
     *
     * In principle, in a partial aggregate, we could share the transition
     * state even if the final function is marked as read-write, because the
     * partial aggregate doesn't execute the final function.  But it's too
     * early to know whether we're going perform a partial aggregate.
     */
    shareable = (aggform->aggfinalmodify != AGGMODIFY_READ_WRITE);
 
    /* get info about the output value's datatype */
    get_typlenbyval(aggref->aggtype,
                    &resulttypeLen,
                    &resulttypeByVal);
 
    /* get initial value */
    textInitVal = SysCacheGetAttr(AGGFNOID, aggTuple,
                                  Anum_pg_aggregate_agginitval,
                                  &initValueIsNull);
    if (initValueIsNull)
        initValue = (Datum) 0;
    else
        initValue = GetAggInitVal(textInitVal, aggtranstype);
 
    ReleaseSysCache(aggTuple);
 
    /*
     * 1. See if this is identical to another aggregate function call that
     * we've seen already.
     */
    aggno = find_compatible_agg(root, aggref, &same_input_transnos);
    if (aggno != -1)
    {
        AggInfo    *agginfo = list_nth_node(AggInfo, root->agginfos, aggno);
 
        agginfo->aggrefs = lappend(agginfo->aggrefs, aggref);
        transno = agginfo->transno;
    }
    else
    {
        AggInfo    *agginfo = makeNode(AggInfo);
 
        agginfo->finalfn_oid = aggfinalfn;
        agginfo->aggrefs = list_make1(aggref);
        agginfo->shareable = shareable;
 
        aggno = list_length(root->agginfos);
        root->agginfos = lappend(root->agginfos, agginfo);
 
        /*
         * Count it, and check for cases requiring ordered input.  Note that
         * ordered-set aggs always have nonempty aggorder.  Any ordered-input
         * case also defeats partial aggregation.
         */
        if (aggref->aggorder != NIL || aggref->aggdistinct != NIL)
        {
            root->numOrderedAggs++;
            root->hasNonPartialAggs = true;
        }
 
        get_typlenbyval(aggtranstype,
                        &transtypeLen,
                        &transtypeByVal);
 
        /*
         * 2. See if this aggregate can share transition state with another
         * aggregate that we've initialized already.
         */
        transno = find_compatible_trans(root, aggref, shareable,
                                        aggtransfn, aggtranstype,
                                        transtypeLen, transtypeByVal,
                                        aggcombinefn,
                                        aggserialfn, aggdeserialfn,
                                        initValue, initValueIsNull,
                                        same_input_transnos);
        if (transno == -1)
        {
            AggTransInfo *transinfo = makeNode(AggTransInfo);
 
            transinfo->args = aggref->args;
            transinfo->aggfilter = aggref->aggfilter;
            transinfo->transfn_oid = aggtransfn;
            transinfo->combinefn_oid = aggcombinefn;
            transinfo->serialfn_oid = aggserialfn;
            transinfo->deserialfn_oid = aggdeserialfn;
            transinfo->aggtranstype = aggtranstype;
            transinfo->aggtranstypmod = aggtranstypmod;
            transinfo->transtypeLen = transtypeLen;
            transinfo->transtypeByVal = transtypeByVal;
            transinfo->aggtransspace = aggtransspace;
            transinfo->initValue = initValue;
            transinfo->initValueIsNull = initValueIsNull;
 
            transno = list_length(root->aggtransinfos);
            root->aggtransinfos = lappend(root->aggtransinfos, transinfo);
 
            /*
             * Check whether partial aggregation is feasible, unless we
             * already found out that we can't do it.
             */
            if (!root->hasNonPartialAggs)
            {
                /*
                 * If there is no combine function, then partial aggregation
                 * is not possible.
                 */
                if (!OidIsValid(transinfo->combinefn_oid))
                    root->hasNonPartialAggs = true;
 
                /*
                 * If we have any aggs with transtype INTERNAL then we must
                 * check whether they have serialization/deserialization
                 * functions; if not, we can't serialize partial-aggregation
                 * results.
                 */
                else if (transinfo->aggtranstype == INTERNALOID &&
                         (!OidIsValid(transinfo->serialfn_oid) ||
                          !OidIsValid(transinfo->deserialfn_oid)))
                    root->hasNonSerialAggs = true;
            }
        }
        agginfo->transno = transno;
    }
 
    /*
     * Fill in the fields in the Aggref (aggtranstype was set above already)
     */
    aggref->aggno = aggno;
    aggref->aggtransno = transno;
}

References Aggref::aggdistinct, AggTransInfo::aggfilter, Aggref::aggfilter, Aggref::aggfnoid, AGGFNOID, PlannerInfo::agginfos, Aggref::agglevelsup, Aggref::aggno, Aggref::aggorder, AggInfo::aggrefs, PlannerInfo::aggtransinfos, Aggref::aggtransno, AggTransInfo::aggtransspace, AggTransInfo::aggtranstype, AggTransInfo::aggtranstypmod, Aggref::aggtype, AggTransInfo::args, Aggref::args, Assert(), AggTransInfo::combinefn_oid, AggTransInfo::deserialfn_oid, elog(), ERROR, TargetEntry::expr, exprType(), exprTypmod(), AggInfo::finalfn_oid, find_compatible_agg(), find_compatible_trans(), FUNC_MAX_ARGS, get_aggregate_argtypes(), get_typlenbyval(), GetAggInitVal(), GETSTRUCT, PlannerInfo::hasNonPartialAggs, PlannerInfo::hasNonSerialAggs, HeapTupleIsValid, initValue(), AggTransInfo::initValueIsNull, lappend(), linitial, list_length(), list_make1, list_nth_node, makeNode, NIL, PlannerInfo::numOrderedAggs, ObjectIdGetDatum(), OidIsValid, ReleaseSysCache(), resolve_aggregate_transtype(), SearchSysCache1(), AggTransInfo::serialfn_oid, AggInfo::shareable, SysCacheGetAttr(), AggTransInfo::transfn_oid, AggInfo::transno, AggTransInfo::transtypeByVal, and AggTransInfo::transtypeLen.

Referenced by preprocess_aggrefs_walker().

preprocess_aggrefs()

void preprocess_aggrefs	(	PlannerInfo *	root,
		Node *	clause
	)

Definition at line 111 of file prepagg.c.

{
    (void) preprocess_aggrefs_walker(clause, root);
}

References preprocess_aggrefs_walker().

Referenced by grouping_planner().

preprocess_aggrefs_walker()

static bool preprocess_aggrefs_walker ( Node *  node, PlannerInfo *  root  )

static

Definition at line 325 of file prepagg.c.

{
    if (node == NULL)
        return false;
    if (IsA(node, Aggref))
    {
        Aggref     *aggref = (Aggref *) node;
 
        preprocess_aggref(aggref, root);
 
        /*
         * We assume that the parser checked that there are no aggregates (of
         * this level anyway) in the aggregated arguments, direct arguments,
         * or filter clause.  Hence, we need not recurse into any of them.
         */
        return false;
    }
    Assert(!IsA(node, SubLink));
    return expression_tree_walker(node, preprocess_aggrefs_walker,
                                  (void *) root);
}

References Assert(), expression_tree_walker, IsA, and preprocess_aggref().

Referenced by preprocess_aggrefs().